Recognition of Indian Sign Language using Hand geometry and Neural Network

Deaf-mute persons communicate using hand gestures, i.e., sign language, making it challenging to share with other persons. A self-effacing Indian sign language recognition system for some real-world gestures is constructed and tested. A dataset consisting of twenty daily-life words used in Indian sign language is created under different variations and lighting conditions. The hand geometrical features are used to estimate hand joints from RGB images. The hand joint coordinates are calculated to determine the distances and angles between the finger’s joints. The proposed model is cost-effective and requires no extra sensors or devices. The experimental results show that the model has a recognizable accuracy of 98.78% on functional gestures.